Open Event Extraction of Emerging Events within Sparse and Fragmented Cybercrime and Technical Online Forums

Novel Natural Language Processing (NLP) approaches will be developed to advance the state of the art of Open Event Extraction in the context of specialist technical online forums (e.g. cybercrime forums) where conversations around new events can be sparse, fragmented and involve out of vocabulary (e.g. new names for cybercrime products). Event schema induction and few/zero shot learning algorithms will be explored. A novel deep learning model for open event extraction will be developed, and the importance of different techniques and parameters (e.g. predefined word embeddings, attention model, few shot algorithm) will be evaluated and compared to well-known benchmark algorithms in the open event extraction literature. Performance across a set of specialist domain datasets (e.g. cybercrime, medical, legal) will be evaluated to explore how results generalize between domains. Opportunities will be available via the supervisor to present results to the National Crime Agency (NCA) as a potential avenue to impact. A Criminology academic has agreed to act as an advisor on cybercrime issues.